Dynamic weather and track conditions?

Smoke&Slide
The frame rate drops in GT5p, were due to not having v-sync Dravonic.Wipeout Fury had that & never had a problem with fps.Also the screen is never tearing.None of the PS3 games, are using all the juice in the PS3 at the mo'.Developers are finding that bit difficult.
Click to expand...

Wipeout can afford to have V-sync and achieves a constant frame rate by never pushing the GPU to the limit. That's how you get no tearing and no frame rate drops.

GT5P on the other hand, by aiming to use the GPU to the limit can't afford to use resources on V-sync and therefore has no V-sync. When it tops out the capacity of the GPU you get frame drops.

Anyway, V-sync is a counter measure for tearing. It doesn't prevent frame rate dropping, it only regulates it to prevent tearing.
 
Last edited:
snoopybg
Wait, I thought weather was confirmed.
Click to expand...

Nothing is confirmed until you have actually seen it in action, in a trailer or on your own PS3.

Latest preview of a magazine in holland said that there was no weather effects, Yamauchi says it could be implemented if it adds something and if it does not take anything away from the full experience, and we have some people and magazines that say it will be in the game.

Who is right? stay tuned for the next episode i guess ;)
 
Dravonic
You would, Kazunori wouldn't. GT has always been about eye candy. You may not agree with it, but that's what it is. It's a characteristic that sets it apart from other sims.
Click to expand...

Unfortunately I think you're right. Don't get me wrong, I love the GT series, have since the first one on PS1 but I can't help thinking about an ideal game I'd like to play and I don't think PD is on the same page as me, despite the fact that they are more than capable of producing it right now.

Personally I'd sacrifice truckloads of polys on the cars to put more detail into the tracks. Imagine what you could do with and extra 4million polys of track detail? That's what you'd save by knocking car detail down by half over 16 vehicles btw.

Since more than half the time I'm playing the game I'm looking at the the track anyway I know where I'd rather see the detail. Spend a million or so on teh scenery and the change on the actual road surface itself and I'm pretty sure the drive would be improved by the same factor it was going from GT4 to GT5p.

The more I hear about how long it took them to model the lights on the new Audi the more I find myself wishing a pretender would appear to challenge PD for the throne. Like I said before - KY has it arse-backwards in a lot of respects.
 
P3nT4gR4m
Unfortunately I think you're right. Don't get me wrong, I love the GT series, have since the first one on PS1 but I can't help thinking about an ideal game I'd like to play and I don't think PD is on the same page as me, despite the fact that they are more than capable of producing it right now.

Personally I'd sacrifice truckloads of polys on the cars to put more detail into the tracks. Imagine what you could do with and extra 4million polys of track detail? That's what you'd save by knocking car detail down by half over 16 vehicles btw.

Since more than half the time I'm playing the game I'm looking at the the track anyway I know where I'd rather see the detail. Spend a million or so on teh scenery and the change on the actual road surface itself and I'm pretty sure the drive would be improved by the same factor it was going from GT4 to GT5p.

The more I hear about how long it took them to model the lights on the new Audi the more I find myself wishing a pretender would appear to challenge PD for the throne. Like I said before - KY has it arse-backwards in a lot of respects.
Click to expand...

Well a pretender already appeared and chances are you know who it is. And like you prefer, they spent more resources on the track. The problem is that trying to top GT will most likely result in failure.

It all comes down to preference and luckily I happen to share Kazunori's. I wish the tracks looked better, but if I had to choose between epic looking cars with passable tracks or OK-ish looking cars with OK-ish looking tracks I would choose the first.
 
P3nT4gR4m
Personally I'd sacrifice truckloads of polys on the cars to put more detail into the tracks. Imagine what you could do with and extra 4million polys of track detail? That's what you'd save by knocking car detail down by half over 16 vehicles btw.
Click to expand...

Exactly. And there would be only marginal difference between 500k and 250k poly cars, like 3%. Still miles better than Forza and others. And for those 5% in car look we all are going to watch those under average tracks full of cardboards and dead computer atmosphere.
 
snicker3s
Explain?
Click to expand...

I think what he is trying to say is, no one can really tell the difference between 500k and 250k polygons in a game. The additional overhead from the detail could have been used for other more detailed aspects of the game, such as the track, trees, etc.
 
I think the poly increase could be because of the damage... I have taken a basic 3D design course and it is better for deformation effects if you have more polygons. For example, if you have a plane that is made of 2 rectangles placed together, it can be manipulated at one vertex. if you add another rectangle in, you have another vertex to manipulate, therefore adding better quality deformation and so on.
 
cool, this discussion has got me thinking of a way of significantly reducing the impact that doubling the number of polys on things like car models would have.
 
Stonemonkey
cool, this discussion has got me thinking of a way of significantly reducing the impact that doubling the number of polys on things like car models would have.
Click to expand...

If it's not too mind bending let us know what it is, you could use it in your "Marshall Hero" game!
 
Ok, a couple of other things too though.

When drawing a 500k poly model there's not that number actually drawn, you only see one side of the car so that alone pretty much halves the number and there are other cases where polys don't have to be drawn.

Also when you increase the model poly count the number of pixels to be filled doesn't change significantly, the things that do change are the geometry calcs, the triangle setups and the memory used by the model.

It has been said in another post that the increase in geometry detail won't give much of an improvement in the image which in part is true, much of the car bodywork is made of smoothly curved panels and while the surface can be drawn with lower poly count and retain the smoothness, the edges and sillouette edges do suffer badly when poly count is lowered.

My idea, which I have no idea if it'd be applicable to the PS3 but is something I could make use of:

Using a low poly mesh but with an extra vert in the middle of each triangle edge, effectively quadrupling the model poly count, if a low poly tri appears on the edge of a surface or sillouette then use the extra verts to sub divide the tri so the majority of the model is drawn as low poly detail but the high poly detail is used where it'll make a difference. This might even result in lowering the total polys needed to draw an object but give better looking results.

Man, that was a long post to make from a phone.
 
Stonemonkey
Ok, a couple of other things too though.

When drawing a 500k poly model there's not that number actually drawn, you only see one side of the car so that alone pretty much halves the number and there are other cases where polys don't have to be drawn.

Also when you increase the model poly count the number of pixels to be filled doesn't change significantly, the things that do change are the geometry calcs, the triangle setups and the memory used by the model.

It has been said in another post that the increase in geometry detail won't give much of an improvement in the image which in part is true, much of the car bodywork is made of smoothly curved panels and while the surface can be drawn with lower poly count and retain the smoothness, the edges and sillouette edges do suffer badly when poly count is lowered.

My idea, which I have no idea if it'd be applicable to the PS3 but is something I could make use of:

Using a low poly mesh but with an extra vert in the middle of each triangle edge, effectively quadrupling the model poly count, if a low poly tri appears on the edge of a surface or sillouette then use the extra verts to sub divide the tri so the majority of the model is drawn as low poly detail but the high poly detail is used where it'll make a difference. This might even result in lowering the total polys needed to draw an object but give better looking results.

Man, that was a long post to make from a phone.
Click to expand...

👍

Well done posting that on a phone...must admit I don't really know what it all means but it sounds good!
 
Imagine drawing a 3D football (soccer for the U.S.) modelled as a triangular mesh. As a low poly model it could look ok except that it might be sort of octagonal in shape around the visible/sillouette edge, a higher poly model has a more rounded edge but the quality otherwise isn't much improved. So use low poly detail in the middle of the visible areas and high poly at the edges.
The models can't be pre made like that though, the football can spin so the code has to decide during rendering whether to use the high or low detail for each poly.

There will be changes in level of detail for cars being further away I'm pretty sure of that but I'd never thought of changing the level in this way before.
 
@ stonemonkey: got ya. That sounds like a pretty good idea, it's the rough edges that ruin the illusion and stops it looking real. Do you know if this method is already in practice? If not you should patent it!
 
evilneal
@ stonemonkey: got ya. That sounds like a pretty good idea, it's the rough edges that ruin the illusion and stops it looking real. Do you know if this method is already in practice? If not you should patent it!
Click to expand...

Doesn't he need to have actual code capable of doing it to patent it? Can you really patent an idea? Anyway, that sort of thing may be used already.
 
Dravonic
Doesn't he need to have actual code capable of doing it to patent it? Can you really patent an idea? Anyway, that sort of thing may be used already.
Click to expand...

Pretty easy to do code-wise, the organisation of the model data could involve quite a bit of work though.

As I said, changes to level of detail is used in most games but usually for distant objects but I've never heard of it being used like what I've thought of here. Could be handy for landscapes too.

EDIT: found something similar called 'progressive LOD' which I imagine would involve what I'm thinking of probably along with some other techniques combined. I could see it being more than likely something like that would be used on the cars in GT5 so that although the models are made from 500k polys there's work done to simplify that before it's drawn and the actual detail levels are dynamic. I'm not saying that's certain but it makes sense to me.
 
Last edited:
Stonemonkey
Ok, a couple of other things too though.

When drawing a 500k poly model there's not that number actually drawn, you only see one side of the car so that alone pretty much halves the number and there are other cases where polys don't have to be drawn.

Also when you increase the model poly count the number of pixels to be filled doesn't change significantly, the things that do change are the geometry calcs, the triangle setups and the memory used by the model.

It has been said in another post that the increase in geometry detail won't give much of an improvement in the image which in part is true, much of the car bodywork is made of smoothly curved panels and while the surface can be drawn with lower poly count and retain the smoothness, the edges and sillouette edges do suffer badly when poly count is lowered.

My idea, which I have no idea if it'd be applicable to the PS3 but is something I could make use of:

Using a low poly mesh but with an extra vert in the middle of each triangle edge, effectively quadrupling the model poly count, if a low poly tri appears on the edge of a surface or sillouette then use the extra verts to sub divide the tri so the majority of the model is drawn as low poly detail but the high poly detail is used where it'll make a difference. This might even result in lowering the total polys needed to draw an object but give better looking results.

Man, that was a long post to make from a phone.
Click to expand...

This technique is being used since the dawn of 3D. Since the beginning there would be less polygons on straight sides and a lot on curved sides. Doesn't work exactly as you explained, but gives better performance.
 
PaMu1337
This technique is being used since the dawn of 3D. Since the beginning there would be less polygons on straight sides and a lot on curved sides. Doesn't work exactly as you explained, but gives better performance.
Click to expand...

I don't think you quite get what I'm meaning, I know for curved surfaces it's better to use more polys than for flat surfaces although too few and lighting can suffer. I'm meaning that using a high poly model, the program simplifies parts of the model appropriately during rendering time.
 
Problem is you're not looking at a silhouette. The way the light interacts wit the facing polys is also important. Shining light off a flat plane will not work the same as a profiled car, reflection, specular and shadow-wise. It's not like the pre-T&L days when this idea would have worked. Just fine - we're ray tracing on the fly nowadays ;)
 
P3nT4gR4m
Problem is you're not looking at a silhouette. The way the light interacts wit the facing polys is also important. Shining light off a flat plane will not work the same as a profiled car, reflection, specular and shadow-wise. It's not like the pre-T&L days when this idea would have worked. Just fine - we're ray tracing on the fly nowadays ;)
Click to expand...

We're not raytracing yet, It'll still be cubemapping and interpolating the light, normal and camera vectors. Probably a lot more accurately now with the shaders but not raytracing, definitely not with 500k poly cars.
 
Stonemonkey
We're not raytracing yet, It'll still be cubemapping and interpolating the light, normal and camera vectors. Probably a lot more accurately now with the shaders but not raytracing, definitely not with 500k poly cars.
Click to expand...


My point is that the facing polys still benefit from higher detail looks-wise. You just can't trick this nowadays. Unlike the normal mapping stuff they're up to with characters. Cars are different - it's all gotta look silky smooth.
 
Also, I am in no way suggesting that the car is made into a flat plane in the shape of the sillouette, that's not what it means in computer graphics.
 
Stonemonkey
Also, I am in no way suggesting that the car is made into a flat plane in the shape of the sillouette, that's not what it means in computer graphics.
Click to expand...

I was using that as an extreme example. I didn't think you meant that at all but any reduction in polycount is going to be a step in that direction with regards how the light moves across the surface. It'll stand out as much as blocky edges in the profile view.
 
P3nT4gR4m
My point is that the facing polys still benefit from higher detail looks-wise. You just can't trick this nowadays. Unlike the normal mapping stuff they're up to with characters. Cars are different - it's all gotta look silky smooth.
Click to expand...

Only up to a certain point, beyond that it won't make much difference, however, the difference is more noticable up to a higher detail level around the sillouette edges.
 
Stonemonkey
Only up to a certain point, beyond that it won't make much difference, however, the difference is more noticable up to a higher detail level around the sillouette edges.
Click to expand...

Not sure I agree with you but I'd be more than happy to be proven wrong. Give it a bash with a simple sphere and see how you get on passing light and shadow across it.
 
P3nT4gR4m
Not sure I agree with you but I'd be more than happy to be proven wrong. Give it a bash with a simple sphere and see how you get on passing light and shadow across it.
Click to expand...

Could take a while to set something up but I'm going to look into it.


http://www.people.fas.harvard.edu/~yzhu/doc/tvcg-paper.pdf

Recent developments in GPU technology have induced atrend in interactive VR software, such as 3D games, toperform shading computation at the pixel-level instead ofthe vertex-level. This provides additional incentives forusing view-dependent continuous LOD algorithms asopposed to static LOD selection. View-dependent remesh-ing allows the silhouette areas to have high detail levelswhile keeping other areas at low detail levels
Click to expand...
 
Interesting stuff:tup:

Do you only need the higher poly treatment at the silhoutte of the car? Im thinking of when you draw a car on a piece of paper with a pencil. Everywhere you draw a line on the paper, you'll probably want higher detail in the 3d world. You will also have to recalculate where these edges/silhouettes are every frame. Or do I see this wrong?

Only way to know for sure if the extra calc is worth the while is probably implementing the bastard :)
 
snicker3s
Interesting stuff:tup:

Do you only need the higher poly treatment at the silhoutte of the car? Im thinking of when you draw a car on a piece of paper with a pencil. Everywhere you draw a line on the paper, you'll probably want higher detail in the 3d world. You will also have to recalculate where these edges/silhouettes are every frame. Or do I see this wrong?

Only way to know for sure if the extra calc is worth the while is probably implementing the bastard :)
Click to expand...

Yeah, I imagine around joins in the bodywork, any hard/sharp(ish) edges too, those polys could be flagged in the model data. You would have to work out which are involved in the silhouette at rendertime though but that can be done as part of the backface culling, an edge shared between two triangles where one triangle is visible to the camera and the other isn't means that edge is part of the silhouette.
 

Latest Posts

Back